CN100452091C - Digital watermark method against geometrical attack based on image characteristic region - Google Patents

Abstract

A digital watermark method resisting against geometric attack includes relating watermark information to image character region, generating local character region with invariance on geometric attack by utilizing most stable Harris character point in image, making said region to be round character region by image normalization, carrying out two-stages of wavelet transforms on these round character regions, selecting revision coefficient regions to embed watermark at the second stage wavelet transform and realizing watermark-embedding. The method for picking up watermark is also disclosed.

Description

Digital watermark method against geometrical attack based on image characteristic region

Technical field

The present invention relates to field of information security technology; a kind of specifically digital watermark method, this method can be resisted translation effectively, rotation; geometric attack and conventional signal Processing attacks such as partial cut can be used for the digital picture in the internet is carried out copyright protection.

Background technology

The develop rapidly of computer technology and Internet technology makes people can realize resource sharing easily.Digital picture, audio frequency, the use of video also greatly facilitates its wide-scale distribution on the internet, has meanwhile also produced the Copyright Protection of a sternness.Digital watermark technology is considered to digital picture is carried out the advantageous methods of copyright protection, and it extracts the proof as the copyright owner by embed certain copyright information in image when dispute over copyright takes place.Existing numerous algorithms propose since the phase at the beginning of the nineties in last century produces.But traditional algorithm mainly concentrates on signal Processing such as the traditional filtering of watermark opposing, noise, compression and attacks.As Guo Baolong, " based on the wavelet field digital watermark method of image target area " patent No. that Guo Lei etc. propose: 03134437.2, by image being carried out wavelet transformation and high frequency coefficient being carried out cluster, obtain the sensation target zone, and watermark information is embedded in this zone exactly.Meanwhile, realize the blind Detecting of watermark by connecting each other of watermark and image target area feature.This method is good for the signal Processing attack performance of routine, but can not resist as geometric attacks such as rotation, convergent-divergent, translation, partial cut.And in actual applications, destroyed the synchronism of watermark detection just because of these geometric attacks, promptly can't determine the position that watermark embeds during watermark detection.Generally speaking, very little geometric attack just might cause the watermark detection failure.Therefore, how to design emphasis and the difficult point that the digital watermarking algorithm that can resist geometric attack is a Recent study.At present, the digital watermarking algorithm of resist geometric attacks mainly is divided into following a few class:

(1) at the transform domain embed watermark that geometric transformation is had unchangeability.The watermark of how much fields of invariants embeds classic methods and is based on rotation, convergent-divergent, and translation has the method for Fourier-plum forests (Fourier-Mellin) conversion of unchangeability, as document J.J.K.

Ruanaidh and T.Pun.Rotation, Scale, and Translation Invariant Spread SpectrumDigital Image watermarking, Signal Processing, vol.66, no.3,1998, pp.303-317. and document C.Y.Lin et al.Rotation, Scale, and Translation Resilient Watermarking of Images, IEEE Trans.ImageProcessing, vol.10, no.5,2001, pp.767-782. is described.The shortcoming of this water mark method maximum based on how much fields of invariants is exactly that to add the quality of watermarking images undesirable.

(2) embedded template, and when watermark detection, utilize template to carry out the estimation of geometric transformation parameter, carry out detecting watermark after the corresponding inverse transformation.Classical algorithm such as document S.Pereira and T.Pun, An Iterative Template-MatchingAlgorithm Using the Chirp-z Transform for Digital Image Watermarking, Pattern Recognition, vol.33, no.1,2000, pp.173-175. is described.Because the method that embeds based on template also needs the extra template that embeds except embed watermark information, therefore also often the quality that adds watermarking images is caused bigger influence, for the attack of attempting to destroy template, the detection of watermark is difficulty comparatively.

(3) embed watermark sequence with cyclophysis.These class methods utilize the autocorrelation performance of watermark to carry out the synchronous of watermark information, but the performance of watermark detection is not very desirable.Classic algorithm such as document D.Delannay and B.Macq, Generalized2D Cyclic Patterns for Secret Watermark Generation, Proc.IEEE Int.Conf.Image Processing (ICIP 2000), vol.II, IEEE Press, 2000, pp.72-79. is described.

(4) utilize the feature of image to carry out the synchronous of geometric attack.Conventional watermarking algorithm is seldom considered the feature of image self in embed watermark, normally with pixel fair plays all in the image, watermark is embedded in the entire image.This class algorithm generally only can be resisted conventional filtering, noise, and signal Processing such as compression of images are attacked, to rotation, convergent-divergent, shearing waits the geometric attack poor-performing, is known as first generation watermark.The thought of second generation watermark proposes (M.Kutter by people such as Kutter, S.K.Bhattacharjee, and T.Ebrahimi, Towards second generation watermarking schemes, inProc.IEEE Int.Conf.Image Processing, 1999, pp.320-323.).Its outstanding feature be exactly effectively utilize features such as unique point in the image, edge carry out watermark information synchronously, this type of algorithm can be resisted geometric attack mostly.Classical algorithm is that people such as Bas is at document P.Bas in the second generation watermark, J.-M.Chassery, and B.Macq, Geometricallyinvariant watermarking using feature points, IEEE Trans.Image Processing.vol.11, no.9,2002, the method that proposes among the pp.1014-1028..The block diagram that its watermark embeds and extracts as depicted in figs. 1 and 2.

Watermark embed process shown in Figure 1 is: at first utilize the unique point of Harris angular-point detection method extraction original image, and the unique point of gained is gathered the triangulation that carries out plane point set, generate a series of triangular image piece.When watermark embeds, at first original watermark signal is converted into size identical with image block and shape by affined transformation, then in conjunction with human visual system's (HVS) visual masking characteristic on the spatial domain is added to watermark signal original picture block, obtain adding the image block of watermark, substitute original image block at last, obtain adding watermarking images.

Watermark extraction process shown in Figure 2 is: at first utilize the Harris angular-point detection method to extract the unique point of image under fire, and to the unique point of gained the triangulation of plane point set is carried out in set, obtain leg-of-mutton image block, each image block is carried out affined transformation it is converted to original watermark have identical size and shape, and carry out Wei Na (Wiener) filtering.Ask before the filtering error image piece of image block behind the image block and filtering, and calculate the related coefficient of itself and original watermark, determine finally that according to the related coefficient of all pieces whether watermark exists.

This algorithm that the people proposed such as Bas can be resisted JPEG (joint photographic experts group) JPEG compression, partial cut, and the low-angle rotation, geometric attacks such as small scale convergent-divergent, but this algorithm has the following disadvantages:

(1) in the characteristic extraction procedure, the unique point number of extraction is too much, and the stability of many unique points is not enough, to such an extent as in the watermark detection process, it is limited to obtain synchronous triangle number preferably.

(2) in watermark embed process, need original watermark is converted to size and the shape identical with the triangular image piece, and when watermark detection, the triangular image piece need be changed into size identical and shape with original watermark, there is interpolation arithmetic in these two processes, cause difference error, influence the detection of watermark.

(3) embedding of watermark signal realizes by it directly is superimposed in the original image respective regions in the spatial domain, so the embedding of watermark is similar to superposition random noise on original image.Therefore, when image is subjected to noise, when filtering etc. are attacked, watermark information can directly be influenced, so this algorithm is relatively poor for this type of attack performance.And rotated at image, during geometric attacks such as convergent-divergent, also can cause direct influence to watermark detection because of the error that interpolation produces.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose a kind of wavelet field digital watermark method, attack, the Internet images is carried out effective copyright protection with opposing geometric attack and conventional signal Processing based on image characteristic region.

The object of the present invention is achieved like this:

The present invention makes full use of the thought of second generation digital watermark, its technical scheme is that watermark information is combined with the invariant feature of image, utilize Harris unique point the most stable in the image to generate geometric attack and conventional signal Processing are attacked the circular feature zone that all has unchangeability, and be rotated normalization, reasonably choose discrete wavelet transform coefficients, realize the embedding and the extraction of watermark in wavelet field.Its concrete grammar comprises watermark embedding and watermark extracting.

Described watermark embeds and comprises the steps:

(1) generates an original watermark sequence w={w by random sequence generator ₁, w ₂, w ₃..., w _n, utilize key that this original watermark sequence is carried out scramble, obtain behind the scramble watermark sequence w '=w ' ₁, w ' ₂, w ' ₃..., w ' _n;

(2) utilize the Harris angular-point detection method to extract the unique point of original image, with each unique point is that circular regions is got at the center, in all border circular areas, the border circular areas of choosing the pairing response of circle centre position unique point and be local maximum is as embedding the candidate feature zone;

(3) to embedding overlapped characteristic area in the candidate feature zone, the response that compares its circle centre position unique point, select a characteristic area of the response maximum of unique point, and remove and there is overlapping characteristic area in this characteristic area, if still there is overlapped situation in remaining characteristic area, aforesaid operations is carried out in circulation, up to the preferred embedding circular feature zone that obtains non-overlapping copies, these are preferably embedded the circular feature zone from original image, take out, and be rotated normalization;

(4) after the normalization each is preferably embedded the circular feature zone and carry out the two-stage wavelet transform, and in the horizontal high-frequency sub-band of second level wavelet transformation and vertical high frequency subband, choose the wavelet coefficient zone C of annular respectively ₁And C ₂, embed the coefficient region that to revise as watermark;

(5) according to the watermark sequence w ' behind the scramble=w ' ₁, w ' ₂, w ' ₃..., w ' _nRevise horizontal high-frequency sub-band coefficient region C respectively ₁With vertical high frequency sub-band coefficients zone C ₂The wavelet coefficient of correspondence position carries out wavelet inverse transformation to amended wavelet coefficient, and the watermark behind the scramble is embedded in each characteristic area;

(6) reverse rotation normalization is carried out in the circular feature zone of described embed watermark, and corresponding zone in the original image that playbacks, the image that view picture adds watermark obtained.

Described watermark extracting comprises the steps:

(1) utilize the Harris angular-point detection method to extract the unique point of image under fire, with each unique point is that circular regions is got at the center, in all border circular areas, the border circular areas of choosing circle centre position unique point corresponding response value and be local maximum is as extracting the candidate feature zone;

(2) to extracting overlapped characteristic area in the candidate feature zone, the response that compares its circle centre position unique point, select a characteristic area of the response maximum of unique point, and remove and there is overlapping characteristic area in this characteristic area, if still there is overlapped situation in remaining characteristic area, aforesaid operations is carried out in circulation, up to the preferred extraction circular feature zone that obtains non-overlapping copies, these are preferably extracted the circular feature zone under fire taking out the image, and be rotated normalization;

(3) after the normalization each is preferably extracted the circular feature zone and carry out the two-stage wavelet transform, and in the horizontal high-frequency sub-band of second level wavelet transformation and vertical high frequency subband, choose the wavelet coefficient zone C of annular respectively ₁And C ₂, the coefficient region that will compare as watermark extracting;

(4) more described coefficient region C ₁With C ₂The wavelet coefficient of correspondence position, extract watermark according to following formula:

Wherein, coef _iBe horizontal high-frequency sub-band coefficient region C ₁Corresponding wavelet coefficient, coef _jBe vertical high frequency sub-band coefficients zone C ₂Corresponding wavelet coefficient;

(5) operate the watermark sequence that is extracted at first by the watermark extracting of described (4) ${\hat{w}}^{\′}=\{{\hat{w}}_1^{\′},{\hat{w}}_2^{\′},{\hat{w}}_3^{\′},\·\·\·,{\hat{w}}_n^{\′}\},$ Utilize key to the unrest that is inverted of the watermark sequence of this extraction, obtain final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\};$

(6) calculate final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\}$ With original watermark sequence w={w ₁, w ₂, w ₃..., w _nNormalized Cross Correlation Function NC, promptly

$\mathrm{NC}=\frac{{\mathrm{\Σ}}_{i=1}^{n}w\left(i\right)\hat{w}\left(i\right)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}^2\left(i\right)}\sqrt{{\mathrm{\Σ}}_{i=1}^n{\hat{w}}^2\left(i\right)}}$

In the formula: w represents original watermark,

The watermark that expression is extracted, n represents the length of watermark sequence;

(7) a NC value and a threshold value T who calculates compared, exists if this NC value, is then judged watermark greater than this threshold value T, on the contrary watermark do not exist, threshold value T is set to 0.8.

Above-mentioned data waterprint embedded method, wherein described in (2) choose circle centre position unique point corresponding response value be the border circular areas of local maximum as embedding the candidate feature zone, carry out as follows:

(1) utilizes the Harris angular-point detection method to extract the unique point of original image, be designated as and embed unique point set Ω ₁

(2) to embed unique point set Ω ₁In each unique point be the center of circle, getting radius is the embedding border circular areas of R;

(3) with the response of other point in the response of described embedding border circular areas circle centre position unique point and the described embedding border circular areas relatively,, then keep this embeddings circular feature zone as embedding candidate feature zone if response is a local maximum.

Above-mentioned data waterprint embedded method, the wherein said wavelet coefficient zone C of choosing in horizontal high-frequency sub-band of second level wavelet transform and the vertical high frequency subband ₁With C ₂The time, get C ₁With C ₂Be the identical annulus of size, the center of circle of these two annulus is positioned at the geometric center of subband separately, inside radius r ₁〉=6 pixels, external radius r ₂Inradius r less than subband ₃

Above-mentioned digital watermarking extracting method, wherein (1) described choose circle centre position unique point corresponding response value be the border circular areas of local maximum as extracting the candidate feature zone, carry out as follows:

(1) utilize the Harris angular-point detection method to extract the unique point of image under fire, be designated as the extract minutiae set omega ' ₁

(2) with the extract minutiae set omega ' ₁In each unique point be the center of circle, getting radius is the extraction border circular areas of R;

(3) with the response of other point in the response of described extraction border circular areas circle centre position unique point and the described extraction border circular areas relatively,, then keep this extractions circular feature zone as extraction candidate feature zone if response is a local maximum.

Above-mentioned digital watermarking extracting method, wherein said in the horizontal high-frequency sub-band and vertical high frequency subband of second level wavelet transformation, the wavelet coefficient zone C of choosing respectively ₁With C ₂Be that two centers of circle are positioned at the annulus of subband geometric center separately, two annulus big or small identical, inside radius r ₁〉=6 pixels, external radius r ₂Inradius r less than subband ₃

The present invention has following effect:

(1) owing to adopt the embedding zone of annular in the wavelet conversion coefficient, watermark information can not cause losing of watermark information because of adding the playback respective regions of original image of watermark characteristic area at last.Because the discreteness of digital picture, the characteristic area that uses in actual watermark embedding and the leaching process is to be inscribed circle with the circular feature zone, the square image blocks of boundary member zero padding.This has just determined watermark to embed optimal conversion is wavelet transform.Because, in discrete Fourier transform (DFT) (DFT) and the wavelet transform (DWT), have only wavelet transform to have localization property in discrete cosine transform (DCT) commonly used.If adopt discrete cosine transform or discrete Fourier transform (DFT), then will influence each pixel in spatial domain by the information of watermark behind the modification frequency coefficient embed watermark, comprise borderline region.And borderline region will be rejected when adding the playback of watermark characteristic area, must cause losing of part watermark information like this.And when utilizing wavelet transform, the wavelet coefficient of modification is positioned at desirable circular feature intra-zone, i.e. annulus C ₁With C ₂Losing of watermark information just can not caused last when adding the playback of watermark characteristic area like this in inside, so the watermark information after the reconstruct only can spread in the desirable circular feature zone, can not spread to boundary member.

(2) owing to adopt the embedding zone of annular in the wavelet conversion coefficient, the embedding of watermark can not influence the primitive characteristics point.The embedding of watermark should influence the primitive characteristics point as small as possible, positions because these unique points will be used to that when watermark extracting watermark is embedded the zone.Circle ring area C ₁With C ₂Interior watermark embeds and makes the watermark information that embeds be dispersed in the annulus of corresponding spatial domain, and the unique point at distance center place has certain distance, therefore can not impact the primitive characteristics point.

(3) because watermark embeds the stability in zone, can guarantee to be rotated at image, translation during geometric attacks such as partial cut, can be extracted watermark effectively; Can resist simultaneously conventional signal Processing and attack, as JPEG (joint photographic experts group) JPEG compression, noise, filtering or the like.

(4), thereby can obtain good picture quality owing to employing embed watermark in the local feature zone of image.

Description of drawings

Fig. 1 embeds block diagram for existing Bas algorithm watermark.

Fig. 2 is existing Bas algorithm watermark extracting block diagram.

Fig. 3 embeds block diagram for watermark of the present invention.

Fig. 4 is a watermark extracting block diagram of the present invention.

Fig. 5 is the generative process figure in circular feature of the present invention zone, wherein 5 (a) are original image, 5 (b) are the primitive character point diagram, 5 (c) are the stable characteristics point diagram in part, 5 (d) are final selected feature point diagram, 5 (e) the embedding candidate feature areal map that local stable characteristics point is determined of serving as reasons, 5 (f) the final selected definite preferred embedding circular feature areal map of unique point of serving as reasons.

Fig. 6 is the normalization figure in circular feature of the present invention zone, wherein 6 (a) are original image circular feature areal map, 6 (b) are the circular feature areal map of 30 ° of image correspondences of rotation, 6 (c) are original image characteristic area normalization figure as a result, and 6 (d) are for rotating 30 ° of image characteristic region figure normalization figure as a result.

Fig. 7 (a) is the desirable circular feature areal map of the present invention.

Fig. 7 (b) is the image block figure of the actual use of the present invention.

The wavelet coefficient areal map that Fig. 8 embeds for watermark of the present invention.

Fig. 9 is that watermark of the present invention embeds instance graph, and wherein Fig. 9 (a) is an original image, and Fig. 9 (b) is preferred circular feature areal map, and 9 (c) are for adding watermarking images, and 9 (d) are error image.

Embodiment

It is following that the present invention is described in further detail with reference to accompanying drawing.

With reference to accompanying drawing 3 and Fig. 5, watermark embed step of the present invention is as follows:

The first step: extract the circular feature zone that watermark embeds.

To original image Fig. 5 (a), utilize Harris angular-point detection method extract minutiae, the unique point of initial extraction is shown in Fig. 5 (b).To piece image I (x, y), the extraction step of Harris unique point is specific as follows:

(1) utilize following formula compute gradient image:

$\left\{\begin{array}{c}X=I\⊗(-\mathrm{1,0,1})=\∂I/\∂X\\ Y=I\⊗{(-\mathrm{1,0,1})}^T=\∂I/\∂Y\end{array}\right.$

Wherein

The expression convolution, X represents the gradient image of horizontal direction, Y represents the gradient image of vertical direction. With

Respectively (x is y) in the gradient of level and vertical direction for presentation video I.

(2) structure autocorrelation matrix:

Order $\left\{\begin{array}{c}A=X^2\⊗z\\ B=Y^2\⊗z\\ C=\left(\mathrm{XY}\right)\⊗z\end{array}\right.$

Z=exp ((X wherein ²+ y ²)/2 σ ²) be Gauss's smoothing windows function.

Autocorrelation matrix then $M=\left[\begin{array}{cc}A& C\\ C& B\end{array}\right].$

(3) extract minutiae:

Order $\left\{\begin{array}{c}{D}_{\mathrm{et}}\left(M\right)=\mathrm{AB}-C^2\\ T_{\mathrm{race}}\left(M\right)=A+B\end{array}\right.$ Then Harris unique point response is: R _H=D _Et(M)-kT _Race ²(M)

Wherein, constant k is got between the 0.04-0.06 usually.With R _HCompare with a threshold value, then regarding as greater than this threshold value is unique point, and this threshold value is according to the unique point number setting that will detect, generally more than or equal to 1000.

Remember that detected Harris embeds feature point set and is combined into Ω ₁, as Fig. 5 (b), the circular feature zone radius that extract is R, R is set to the 40-50 pixel among the present invention, with Ω ₁In each unique point p _iBe the center of circle, get the border circular areas that its radius is R, and judge circle centre position unique point, i.e. p _iWhether the response at place is the local maximum in this zone, if keeping characteristics point p then _iAnd determined border circular areas is as embedding the candidate feature zone; Otherwise, give up.

To be expressed as Ω through the candidate feature point set that aforesaid operations obtains ₂, as Fig. 5 (c), this set omega ₂In any one unique point all be the stable characteristics point in part in the image, these features are subjected to most possibly being detected again when geometric attack and conventional signal Processing are attacked at image, and the position that has embedded watermark is positioned.By Ω ₂In the candidate feature zone determined of unique point shown in Fig. 5 (e), as seen, only there is stable characteristics point in the inside in each candidate feature zone.

In Fig. 5 (e), set omega ₂In the candidate feature zone determined of unique point have overlapped situation, in order to obtain the characteristic area of non-overlapping copies, need be to Ω ₂In unique point screen.For this reason, at first from Ω ₂In take out the unique point of response maximum, and remove with the determined characteristic area of this unique point and have unique point in the overlapping characteristic area.If still there is overlapped situation in remaining characteristic area, carry out aforesaid operations to there being overlapping characteristic area circulation, up to the characteristic area that obtains non-overlapping copies, remember that simultaneously final feature point set is combined into Ω ₃, shown in Fig. 5 (d).By Ω ₃The preferred embedding characteristic area that middle unique point is determined is shown in Fig. 5 (f), and the preferred embedding characteristic area non-overlapping copies among Fig. 5 (f) has unchangeability to geometric attack and signal Processing processing, is the invariant features zone, i.e. watermark embeds the zone.

Second step: preferred embedding characteristic area is rotated normalization.

When there was rotation in image, though from original image with to rotate the circular characteristic area content that extracts the later image identical, their direction difference was shown in Fig. 6 (a) and Fig. 6 (b).Therefore, utilize the method for wavelet embed watermark must eliminate the influence that rotation brings.Utilize the image normalization technology based on geometric moment among the present invention, the characteristic area that has rotation is carried out normalization, it is as follows to rotate normalized concrete grammar:

(x, y), its p+q rank geometric moment is to a width of cloth digital picture I

m _p，q＝∫∫ _rx ^py ^qI(x，y)dxdy

The field of definition of Г presentation video wherein, corresponding central moment is

μ _p，q＝∫∫ _r(x-x) ^p(y-y) ^qI(x，y)dxdy

X=m wherein _1,0/ m _0,0, y=m _0,1/ m _0,0The barycenter of presentation video.

Define two tensor t ¹And t ²As follows:

t ¹＝μ _1，2+μ _3，0，t ²＝μ _2，1+μ _0，3

Then rotating normalization angle θ is:

θ＝arctan(-t ¹/t ²)

Rotation normalization angle be with respect to image barycenter (x, y).

Obviously, there are two in following formula to be separated, for example φ and φ+π.In order to obtain a unique normalization angle, definition another one tensor t ³:

t ³＝-t ¹sinφ+t ²cosφ

By making t ³Can determine a unique normalization angle φ for＞0.If t ³＜0, then make φ=φ+π.

Normalization angle φ has been arranged, and only needing with the barycenter is that benchmark rotates the φ angle with image, promptly can obtain normalized image, and normalized image has unchangeability to rotation.Image after the normalization of characteristic area is shown in Fig. 6 (c) and Fig. 6 (d).

The 3rd step: the normalization characteristic area is carried out wavelet transform, determine the coefficient region that watermark embeds.

The embedded location of watermark under the geometric distortion is located by above-mentioned second step, on this basis, just can adopt comparatively the transform domain watermarking algorithm of maturation.For opposing rotation effectively, translation, geometric attacks such as shearing, the present invention is embed watermark in the local invariant characteristic area of image.But in the implementation process of reality, what we can use is not desirable circular feature zone, but is the square image blocks of inscribed circle with the characteristic area, is about to the boundary member zero padding, shown in Fig. 7 (b).In Fig. 7 (b), the corresponding region of original image because the characteristic area that needs after watermark embeds to contain watermark playbacks, and the borderline region of black will be removed at this moment, therefore must guarantee that watermark information is embedded in the desirable circular feature zone, be distributed to the borderline region of black as few as possible, this conversion of adopting with regard to requiring has localization property.Meanwhile, the embedding of watermark will influence near the pixel the primitive character point as small as possible, these unique points can be detected well watermark embedding zone is positioned in the time of could guaranteeing watermark detection like this, and the therefore same conversion that requires to be adopted has localization property.According to above two requirements, the present invention selects wavelet transform (DWT) embed watermark for use.When utilizing the DWT embed watermark, as long as revise wavelet coefficient corresponding in the desirable circular feature zone, then watermark information only can influence the spatial domain pixel of relevant position, and is very little to the influence of borderline region pixel.And in order to influence the primitive characteristics point as small as possible, near the wavelet coefficient the unique point can not be modified.Therefore, the wavelet field watermark embedding coefficient of the present invention's design is selected form as shown in Figure 8 for use.Among Fig. 8, low frequency sub-band LL _2.0With 6 high-frequency sub-band HL _1.1, HH _1.2, LH _1.3, HL _2.1, HH _2.2, LH _2.3Be the wavelet conversion coefficient that original image is carried out obtain behind the two-stage wavelet transform, wherein HL _1.1, HH _1.2, LH _1.3Be respectively horizontal high frequency, diagonal line high frequency and vertical high frequency subband: the HL of first order wavelet transformation _2.1, HH _2.2, LH _2.3Be respectively horizontal high frequency, diagonal line high frequency and the vertical high frequency subband of second level wavelet transformation.Horizontal high-frequency sub-band HL at second level wavelet transformation _2.1With vertical high frequency subband LH _2.3In, choose the coefficient region of two annulars respectively, the circle ring area note in the wherein horizontal high-frequency sub-band is made C ₁, the circle ring area note in the vertical high frequency subband is made C ₂These two annulus C ₁With C ₂The center of circle lay respectively at horizontal high-frequency sub-band HL _2.1With vertical high frequency subband LH _2.3Geometric center, inside radius r ₁〉=6 pixels, external radius r ₂Inradius r less than subband ₃C ₁With C ₂Be the wavelet coefficient zone that the watermark embedding will be revised among the present invention.Watermark information is embedded in this zone can guarantee that the embedding of watermark can not influence the primitive character point, can guarantee that again watermark information concentrates in the desirable characteristic area, and can not cause losing of watermark information because of containing the playback of watermark characteristic area at last.

The 4th step: watermark embeds.

Generate an original watermark sequence w={w by random sequence generator ₁, w ₂, w ₃..., w _nAs watermark sequence to be embedded,, at first use key that original watermark sequence w is carried out scramble for the security of enhanced system, obtain behind the scramble watermark sequence w '=w ' ₁, w ' ₂, w ' ₃..., w ' _n.If coef _iAnd coef _jRepresent that respectively watermark shown in Figure 8 embeds zone C ₁With C ₂The wavelet conversion coefficient of correspondence position embeds the watermark sequence behind the scramble in the characteristic area according to the following procedure:

At first, wavelet coefficient is made amendment embed watermark.If w ' _i=1 and D=coef _i-coef _j＜TH then increases horizontal high-frequency sub-band watermark by following formula and embeds zone C ₁Corresponding wavelet conversion coefficient coef _iValue, reduce the vertical high frequency subband simultaneously and embed zone C ₂Corresponding wavelet conversion coefficient coef _jValue

$\left\{\begin{array}{c}{\mathrm{coef}}_i^{\′}={\mathrm{coef}}_i+(\mathrm{TH}-D)/2\\ {\mathrm{coef}}_j^{\′}={\mathrm{coef}}_j-(\mathrm{TH}-D)/2\end{array}\right.$

If D=coef _i-coef _j〉=TH does not do any operation;

If w ' _i=0 and D=coef _j-coef _i＜TH, carry out following operation:

$\left\{\begin{array}{c}{\mathrm{coef}}_i^{\′}={\mathrm{coef}}_i-(\mathrm{TH}-D)/2\\ {\mathrm{coef}}_j^{\′}={\mathrm{coef}}_j+(\mathrm{TH}-D)/2\end{array}\right.$

If D=coef _j-coef _i〉=TH does not do any operation.Wherein TH is the threshold value of control watermark invisibility and robustness, and this threshold value TH chooses 0.02 among the present invention.

Then, amended wavelet coefficient is carried out wavelet inverse transformation, obtain adding the characteristic area of watermark.

The 5th step: the characteristic area that will add after the watermark carries out reverse rotation normalization, and the relevant position in the original image that playbacks, and obtains the image that view picture adds watermark.

With reference to Fig. 4, watermark extracting of the present invention is carried out according to following process:

The first step: extract the circular feature zone that has embedded watermark in the image under fire, the concrete operations in this step are identical with the leaching process in circular feature zone during watermark embeds.

Second step: the circular feature zone of extracting is rotated normalization, and the concrete operations in this step are identical with the rotation normalization in circular feature zone in the watermark embed process.

By above two the step to the rotation, translation, the zone that has embedded watermark under the geometric attacks such as shearing positions.

The 3rd step: the two-stage wavelet transform is carried out in each normalization circular feature zone, and the selection annular wavelet conversion coefficient zone identical with telescopiny, establish coef _iAnd coef _jBe respectively coefficient region C shown in Figure 8 ₁With C ₂The wavelet coefficient of correspondence position, the extraction of watermark is finished by following equation:

To the watermark of extracting ${\hat{w}}^{\′}=\{{\hat{w}}_1^{\′},{\hat{w}}_2^{\′},{\hat{w}}_3^{\′},\·\·\·,{\hat{w}}_n^{\′}\}$ Utilize the unrest that is inverted of identical key, obtain final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\}.$

The 4th step: calculate final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\}$ With original watermark sequence w={w ₁, w ₂, w ₃..., w _nNormalized Cross Correlation Function NC, that is:

$\mathrm{NC}=\frac{{\mathrm{\Σ}}_{i=1}^{n}w\left(i\right)\hat{w}\left(i\right)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}^2\left(i\right)}\sqrt{{\mathrm{\Σ}}_{i=1}^n{\hat{w}}^2\left(i\right)}}$

In the formula: w represents original watermark,

The watermark that expression is extracted, n represents the length of watermark sequence;

The 5th step: the NC value and the threshold value T that calculate are compared, if this NC value, is then judged watermark existence greater than this threshold value T, on the contrary watermark do not exist, thresholding T is set to 0.8.

Figure 9 shows that watermark of the present invention embeds example.9 (a) represent original image, the circular feature zone that 9 (b) expression watermark embeds, and 9 (c) expression adds watermarking images, and 9 (d) expression contains the error image of watermarking images and original image.Add watermarking images Y-PSNR PSNR value and be respectively 39.41dB and 37.90dB.As can be seen, the present invention adds watermarking images and has very high quality owing to embed watermark in the regional area of image.

Claims (6)

1. the wavelet field resist geometric attacks data waterprint embedded method based on image characteristic region comprises the steps:

(1) generates an original watermark sequence w={w by random sequence generator ₁, w ₂, w ₃..., w _n, utilize key that this original watermark sequence is carried out scramble, obtain behind the scramble watermark sequence w '=w ' ₁, w ' ₂, w ' ₃..., w ' _n;

(2) utilize the Harris angular-point detection method to extract the unique point of original image, with each unique point is that circular regions is got at the center, in all border circular areas, the border circular areas of choosing the pairing response of circle centre position unique point and be local maximum is as embedding the candidate feature zone;

(3) to embedding overlapped characteristic area in the candidate feature zone, the response that compares its circle centre position unique point, select a characteristic area of the response maximum of unique point, and remove and there is overlapping characteristic area in this characteristic area, if still there is overlapped situation in remaining characteristic area, aforesaid operations is carried out in circulation, up to the preferred embedding circular feature zone that obtains non-overlapping copies, these are preferably embedded the circular feature zone from original image, take out, and be rotated normalization;

(4) after the normalization each is preferably embedded the circular feature zone and carry out the two-stage wavelet transform, and in the horizontal high-frequency sub-band of second level wavelet transformation and vertical high frequency subband, choose the wavelet coefficient zone C of annular respectively ₁And C ₂, embed the coefficient region that to revise as watermark;

(5) according to the watermark sequence w ' behind the scramble=w ' ₁, w ' ₂, ' ₃..., w ' _nRevise horizontal high-frequency sub-band coefficient region C respectively ₁With vertical high frequency sub-band coefficients zone C ₂The wavelet coefficient of correspondence position carries out wavelet inverse transformation to amended wavelet coefficient, and the watermark behind the scramble is embedded in each characteristic area;

(6) reverse rotation normalization is carried out in the circular feature zone of described embed watermark, and corresponding zone in the original image that playbacks, the image that view picture adds watermark obtained.

2. the wavelet field resist geometric attacks data waterprint embedded method based on image characteristic region according to claim 1, it is characterized in that described in (2) choose circle centre position unique point corresponding response value be the border circular areas of local maximum as embedding the candidate feature zone, carry out as follows:

(1) utilizes the Harris angular-point detection method to extract the unique point of original image, be designated as and embed unique point set Ω ₁

(2) to embed unique point set Ω ₁In each unique point be the center of circle, getting radius is the embedding border circular areas of R;

(3) with the response of other point in the response of described embedding border circular areas circle centre position unique point and the described embedding border circular areas relatively,, then keep this embeddings circular feature zone as embedding candidate feature zone if response is a local maximum.

3. the wavelet field resist geometric attacks data waterprint embedded method based on image characteristic region according to claim 1 is characterized in that choosing the wavelet coefficient zone C in horizontal high-frequency sub-band of second level wavelet transform and the vertical high frequency subband ₁With C ₂The time, get C ₁With C ₂Be the identical annulus of size, the center of circle of these two annulus is positioned at the geometric center of subband separately, inside radius r ₁〉=6 pixels, external radius r ₂Inradius r less than subband ₃

4. the wavelet field resist geometric attacks digital watermarking extracting method based on image characteristic region comprises the steps:

(1) utilize the Harris angular-point detection method to extract the unique point of image under fire, with each unique point is that circular regions is got at the center, in all border circular areas, the border circular areas of choosing circle centre position unique point corresponding response value and be local maximum is as extracting the candidate feature zone;

(2) to extracting overlapped characteristic area in the candidate feature zone, the response that compares its circle centre position unique point, select a characteristic area of the response maximum of unique point, and remove and there is overlapping characteristic area in this characteristic area, if still there is overlapped situation in remaining characteristic area, aforesaid operations is carried out in circulation, up to the preferred extraction circular feature zone that obtains non-overlapping copies, these are preferably extracted the circular feature zone under fire taking out the image, and be rotated normalization;

(3) after the normalization each is preferably extracted the circular feature zone and carry out the two-stage wavelet transform, and in the horizontal high-frequency sub-band of second level wavelet transformation and vertical high frequency subband, choose the wavelet coefficient zone C of annular respectively ₁And C ₂, the coefficient region that will compare as watermark extracting;

(4) more described coefficient region C ₁With C ₂The wavelet coefficient of correspondence position, extract watermark according to following formula:

Wherein, coef _iBe horizontal high-frequency sub-band coefficient region C ₁Corresponding wavelet coefficient, coef _jBe vertical high frequency sub-band coefficients zone C ₂Corresponding wavelet coefficient;

(5) operate the watermark sequence that is extracted at first by the watermark extracting of described (4) ${\hat{w}}^{\′}=\{{\hat{w}}_1^{\′},{\hat{w}}_2^{\′},{\hat{w}}_3^{\′},\·\·\·,{\hat{w}}_n^{\′}\},$ Utilize key to the unrest that is inverted of the watermark sequence of this extraction, obtain final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\};$

(6) calculate final watermark sequence $\hat{w}=\{{\hat{w}}_1,{\hat{w}}_2,{\hat{w}}_3,\·\·\·,{\hat{w}}_n\}$ With original watermark sequence w={w ₁, w ₂, w ₃..., w _nNormalized Cross Correlation Function NC, promptly

$\mathrm{NC}=\frac{{\mathrm{\Σ}}_{i=1}^{n}w\left(i\right)\hat{w}\left(i\right)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{w}^2\left(i\right)}\sqrt{{\mathrm{\Σ}}_{i=1}^n{\hat{w}}^2\left(i\right)}}$

In the formula: w represents original watermark,

The watermark that expression is extracted, n represents the length of watermark sequence;

(7) a NC value and a threshold value T who calculates compared, exists if this NC value, is then judged watermark greater than this threshold value T, on the contrary watermark do not exist, threshold value T is set to 0.8.

5, the wavelet field resist geometric attacks digital watermarking extracting method based on image characteristic region according to claim 2, it is characterized in that step (1) described choose circle centre position unique point corresponding response value be the border circular areas of local maximum as extracting the candidate feature zone, carry out as follows:

(1) utilize the Harris angular-point detection method to extract the unique point of image under fire, be designated as the extract minutiae set omega ' ₁

(2) with the extract minutiae set omega ' ₁In each unique point be the center of circle, getting radius is the extraction border circular areas of R;

(3) with the response of other point in the response of described extraction border circular areas circle centre position unique point and the described extraction border circular areas relatively,, then keep this extractions circular feature zone as extraction candidate feature zone if response is a local maximum.

6, the wavelet field resist geometric attacks digital watermarking extracting method based on image characteristic region according to claim 2, when it is characterized in that determining wavelet coefficient that watermark extracting adopted, in the horizontal high-frequency sub-band and vertical high frequency subband of second level wavelet transformation, the wavelet coefficient zone C of choosing respectively ₁With C ₂Be that two centers of circle are positioned at the annulus of subband geometric center separately, two annulus big or small identical, inside radius r ₁〉=6 pixels, external radius r ₂Inradius r less than subband ₃

Images